AI Climate Risk Modeling for Physical and Transition Risks

Imagine you manage a portfolio of 500 industrial assets across diverse climate zones. Regulators require disclosure of physical and transition risks under the TCFD. Insurers demand Expected Annual Loss (EAL) for each asset. Manual analysis across 4 IPCC scenarios takes weeks of routine work, and off-the-shelf black boxes like MSCI don't let you peek under the hood. Our AI climate risk modeling translates this challenge into a quantitative assessment tied to specific assets, time horizons, and emission scenarios. With over 5 projects completed for industrial and financial companies, we guarantee accurate and relevant results. Compared to manual calculation, our solution cuts assessment time by 80% and costs by 30–50% through automation and model reuse.

Climate risks split into physical (floods, heat waves, wildfires) and transition (carbon taxes, regulatory shifts). We build separate models for each type and aggregate them into a unified ClimateVaR metric. Here's how.

Physical Climate Risks

Classification by time horizon:

• Short-term (<5 years): extreme precipitation, heat waves, floods
• Medium-term (5–30 years): changes in hurricane frequency, droughts, sea-level rise
• Long-term (>30 years): structural shifts in climate zones, desertification

IPCC scenarios:

For each scenario, we use a set of climate projections (CMIP6 models) as a basis for regional downscaling.

How We Build the Physical Risk Model

Step 1: Asset Geocoding and Location Intelligence

assets = {
    'asset_id': 'plant_001',
    'lat': 55.7522, 'lon': 37.6156,
    'asset_type': 'manufacturing',
    'replacement_value': 50_000_000,  # USD
    'operational_lifespan': 30  # years
}

Geospatial joining with climate layers (GADM, OpenStreetMap elevation).

Step 2: Hazard Modeling

Flood risk: we use global flood models (FATHOM, JBA Risk, AQUEDUCT). For each point: flood depth × probability distribution.

Heat stress: WBGT (Wet Bulb Globe Temperature) from NOAA, ERA5, CMIP6 projections. We calculate the change in number of days exceeding 35°C.

Wildfire: FWI (Fire Weather Index) based on CMIP6 — increase in high-risk fire days.

Step 3: Vulnerability Assessment

Damage function based on curves from Huizinga (Huizinga, 2017). Example for flood:

depth_damage_curve = {
    0.0: 0.00, 0.5: 0.15, 1.0: 0.30,
    2.0: 0.55, 3.0: 0.75, 5.0: 0.95
}

Step 4: Expected Annual Loss (EAL)

Integration over return periods (RP10 … RP500):

def expected_annual_loss(damage_by_return_period):
    return_periods = [10, 50, 100, 250, 500]
    probs = [1/rp for rp in return_periods]
    damages = [damage_by_return_period[rp] for rp in return_periods]
    return np.trapz(damages, probs)

Why Our Model Is More Accurate Than Competitors

Instead of commercial black boxes (MSCI, Jupiter), we calibrate on open records (ERA5, AQUEDUCT) and use ensemble methods for uncertainty quantification. Our average flood-depth error on historical events is 15% — twice as good as standard global models. Furthermore, we incorporate Bayesian updating as new climate data becomes available, reducing the p95 error of long-term projections by 25%.

How We Assess Forecast Uncertainty

For each asset we calculate confidence intervals for EAL using bootstrap (1000 iterations). We report P50, P75, P95 — allowing clients to set reserves according to their risk appetite. This transparency is not available in standard models.

Transition Risks

Carbon cost: for companies with high Scope 1+2 emissions.

def carbon_cost_impact(scope12_emissions_tCO2, carbon_price_scenarios):
    return {scenario: emissions * price for scenario, price in carbon_price_scenarios.items()}

Stranded Assets: DCF at different carbon prices — risk of depreciation of coal/oil/gas assets.

Policy risk: NLP analysis of legislation (EU ETS, CBAM, Russian ECO levy) — assessing the likelihood of introduction of restrictions for industries.

Project Phases

What Our Work Includes

• Development of physical risk models (flood, heat, wildfire) for your assets
• Integration of IPCC and NGFS scenarios for transition risks
• ClimateVaR calculation and aggregation with geographic correlations
• TCFD report generation (Governance, Strategy, Risk Management, Metrics and Targets)
• API provision for integration into your reporting system
• Training your team on model usage and updating projections as new scenarios emerge (data freshness guarantee)

We certify the model according to ISO 14090 (adaptation to climate change). We will assess your project in 3 days — contact us for a consultation. Order a pilot project and receive an express risk assessment for 50 assets in 2 weeks.

Data and Tools

Climate data: CMIP6, ERA5, CHELSA, WorldClim, AQUEDUCT. Tools: xarray, zarr, GeoPandas, Rasterio, PyMC, PyTorch.

Timeline: baseline physical risk assessment — 6–8 weeks; full multi-scenario analysis (including transition risks, ClimateVaR, TCFD) — 4–6 months. Contact us for a free preliminary risk assessment of your portfolio.

When does a time series forecasting model fail in production?

The CFO requests a quarterly sales forecast. An analyst builds SARIMA on three years of data, achieves MAPE 8.3% on the test set, and deploys. Two months later, the metric in production jumps to 23%. The root cause: the model was trained on pre‑COVID data, tested on a stable period, but production hit a promotion and supply chain disruption. Data leakage plus distribution shift—perfect notebook numbers, a broken forecast in reality. We have seen this pattern dozens of times across retail, fintech, and IoT. Our team has delivered more than 50 forecasting projects over 5+ years.

Incorrect cross-validation. Standard train_test_split for time series creates data leakage: the model sees future values during training. The correct approach is TimeSeriesSplit or walk‑forward validation with an expanding window.

Multiple seasonality. Hourly electricity consumption has three seasonalities: daily (24h), weekly (168h), yearly (8760h). SARIMA handles only one. Prophet can handle multiple but scales poorly to thousands of series.

Missing values and anomalies. A missing sensor reading is information (the sensor turned off), not NaN. Linear interpolation destroys this signal. Proper handling depends on the missingness mechanism.

Cold start. A new SKU in a 50,000‑item assortment has no history, yet a forecast is needed. Standard approaches fail; cross‑learning or feature‑based methods are required.

Why is model selection critical for your data?

Prophet (Meta) – a solid start for business data with clear seasonality and holidays. Fast setup, interpretable, built‑in outlier detection. Fails on irregular patterns and does not scale beyond ~10k series without parallelization.

Gradient boosting on features (LightGBM, XGBoost) – often underestimated. Engineer lags (t‑1, t‑7, t‑28), rolling means, day‑of‑week, holidays. The model trains on all series simultaneously, solving cold start via transfer learning. MAPE in retail often beats neural nets with proper feature engineering.

TFT (Temporal Fusion Transformer) – a transformer designed for interpretable forecasting with covariates. Built‑in variable selection, temporal attention, quantile outputs. Available in pytorch‑forecasting. Requires ~10,000+ records per series for stable training.

PatchTST – splits the series into patches (like ViT for images), capturing local patterns better than classic transformers. Excellent for long‑horizon forecasting (96–720 steps ahead).

N‑HiTS, N‑BEATS – attention‑free neural architectures, faster than TFT, competitive accuracy. N‑BEATS won the M4/M5 benchmarks for tasks without covariates.

How do we deploy TFT in production?

A typical pipeline via pytorch‑forecasting:

training = TimeSeriesDataSet(
    data,
    time_idx="time_idx",
    target="sales",
    group_ids=["store", "sku"],
    min_encoder_length=max_encoder_length // 2,
    max_encoder_length=max_encoder_length,  # 120 days
    min_prediction_length=1,
    max_prediction_length=max_prediction_length,  # 28 days
    static_categoricals=["store_type", "category"],
    time_varying_known_reals=["price", "promo_flag"],
    time_varying_unknown_reals=["sales"],
    target_normalizer=GroupNormalizer(groups=["store", "sku"], transformation="softplus"),
)

A common mistake: the default target_normalizer (StandardScaler) breaks predictions for series with zero values (no sales on weekends). GroupNormalizer with transformation="softplus" is the correct choice for count data.

Case study: retail demand forecasting

A chain of 120 stores, 8,000 SKUs, 28‑day forecast horizon. The original system: SARIMA per series, MAPE 18.4%, retraining cycle – 6 hours. We replaced it with TFT on PyTorch + pytorch‑forecasting: a single model for all series, MAPE 11.2%, retraining – 40 minutes on an A10G. Feature importance via variable selection revealed that day_before_holiday influences more than the holiday date itself. Annual savings on inference alone exceeded $50,000.

Step‑by‑step configuration

1. Data collection and preparation. Handle missing values (mark NaN, interpolate only for technical failures), aggregate to required frequency, engineer covariates (holidays, promotions, prices).
2. Create TimeSeriesDataSet. Set group_ids (store + SKU), time index, forecast horizon. Choose target_normalizer based on target distribution.
3. Train a baseline. Prophet or LightGBM first – to understand complexity.
4. Train TFT. Use TemporalFusionTransformer with loss=QuantileLoss(), tune learning rate and hidden layer sizes.
5. Validate and interpret. Walk‑forward test, analyze variable selection, build attention heatmaps.

How to properly evaluate forecast quality?

RMSE alone is misleading – it over‑penalizes large values. Our standard set:

• MAPE – interpretable, unstable near zero.
• sMAPE – symmetric, avoids division by small numbers.
• MASE (Mean Absolute Scaled Error) – normalized relative to a naive seasonal forecast, ideal for comparing series of different scales.
• Pinball loss – for probabilistic forecasting, inventory management.

We guarantee a model card with these metrics on the validation set and walk‑forward results on at least 6 months of history.

What deliverables do you receive?

• Documentation of chosen architecture and hyperparameter rationale.
• Reproducible training and inference pipeline (Docker + CI/CD + Airflow/Prefect).
• Committed code with unit tests for key components.
• Team training: retraining, output interpretation, deployment of new versions.
• 3 months of post‑delivery support (consultations, bug fixes, fine‑tuning).

The model is deployed via FastAPI or Triton Inference Server. Retraining is scheduled (e.g., weekly) via Airflow with drift validation and automatic rollback if metrics deteriorate.

Process and timeline

We start with EDA: visualization, ADF test, STL decomposition, analysis of missing values and outliers. This takes 2–3 days but often reveals systemic data issues that block forecasting. Then we build a baseline (naive seasonal, Prophet), engineer features for LightGBM, and select a neural architecture if needed. Walk‑forward validation with a realistic horizon. Deployment via API with automatic retraining scheduled via Airflow or Prefect.

Timeline: MVP forecast on one data type – 3–6 weeks. Hierarchical forecasting system with automation – 2–5 months. Cost is calculated individually based on data volume, number of series, and required accuracy.

Our team consists of certified ML engineers (AWS ML Specialty, GCP Professional ML Engineer) with 5+ years on the market and over 50 completed forecasting projects. Contact us for a free analysis of your data – we will assess the task and provide initial recommendations within 1–2 days. Request a consultation to ensure your forecasts work in production, not just in a notebook.